Getting to production: The secrets to secure, scalable and cost-effective enterprise AI

In a sign of how quickly enterprises are moving to embrace AI, 70% have moved past the pilot stage and are preparing to release new use cases into the real world, according to a global survey from Economist Impact.

The shift to production means hundreds, potentially thousands, more users will interact with the underlying AI system. Amid this often rapid scale-out, businesses are struggling to maintain reliability and performance. Ensuring unified governance across the massive underlying datasets remains a critical roadblock. In fact, just 29% of practitioners believed their company’s AI investments were production-ready.

In a recent webinar, Databricks CIO Naveen Zutshi and Economist Impact Editorial Director Tamzin Booth outlined the steps that technology leaders can take to securely and cost-effectively build and scale new AI workloads:

• Treat it as a software development project: According to the survey, just 22% of respondents were confident in the ability of their existing IT infrastructure to support new AI workloads. IT leaders must invest in the right underlying platform to support their engineers. Businesses need to unify their data into one location, adopt a single governance catalog, then give developers the tools to build directly on those assets, as well as monitor the resulting workloads.
• Encourage governance and security: Establishing unified governance across all of their company’s data is a key hurdle to deploying and scaling enterprise AI workloads, according to 33% of respondents. However, with data consolidated to a single location, IT leaders can deploy a unified governance catalog across all the enterprise’s assets, allowing users to more confidently build with the information. Companies must also put guardrails in place to let engineers safely and securely unleash their creativity. Technologists are going to use new AI tools regardless. It’s up to IT leaders to make sure it’s being done in a safe and governed manner.
• Focus on the user interface: It’s not easy to get employees to change their workflows. Choosing the right way to incorporate AI into existing processes is critical. At Databricks, when we were building our internal AI agent to provide field staff automated data intelligence, there were a lot of false starts before we found the best way to incorporate the feature. We eventually embedded into existing programs our employees already use. Now, it seamlessly provides the intelligence without the user having to do anything.
• Pick the right AI use case: Organizations typically begin their AI journey with internal use cases so that teams can be confident in their project’s outcomes and reliability. Don’t pursue tech just for tech’s sake - it has to be tied to value and impact. IT leaders must find ways to work closely with users to evangelize and scale the technology, like business hackathons.
• Find your edge: Success in AI all comes back to the data. Instead of one monolithic model, companies are building AI agent systems that take advantage of several models to amplify the end performance. Now, the true competitive advantage comes from how well companies can use proprietary assets to customize these systems and develop unique products. In fact, 66% of global organizations said they recognize the potential in integrating GenAI with proprietary data. And businesses like FactSet and Replit are already generating revenue from AI products built on their own data using agentic systems.
• Be patient with progress: During periods of major transformation, year-over-year change can seem slow. But the progress has been significant. It took businesses over 10 years to infuse the Internet into their operations. Some CIOs aren’t even expecting to reap their AI returns for three years. But as the number of high-quality models goes up, and the price of compute and inference costs continues to drop, there will be a rapid increase in AI Agent Systems that will greatly accelerate the pace of change.

By taking these steps, CIOs help their organizations to use AI to both improve productivity and efficiency, as well as unlock the real prize: business innovation and revenue generation.

To dive deeper into these best practices, watch the webinar: Unlocking Enterprise AI. Or read the full report and its findings here.